Training of neural network for pattern classification using fireworks algorithm

The challenge of training the artificial neural networks (ANNs) which is frequently used for classification purpose has been consistently growing over the last few years, this is probably due to the high dimensional and multi-modal nature of the search space. Nature-inspired metaheuristic algorithms have been successfully employed in the process of weight training of such complex continuous optimization problems. In this paper, a recently proposed fireworks algorithm (FWA) is presented for the training of the parameters of the ANNs. FWA is a class of population-based search method which imitates the explosion process of real fireworks at night. In order to investigate the performance of the proposed method, experiments were conducted on seven benchmark problem instance from the UCI machine learning laboratory and the results obtained by the proposed method are compared with those obtained by krill herd algorithm, harmony search algorithm and genetic algorithm. The results of the evaluation showed superiority of the proposed algorithm in both SSE and training CA and had comparative performance in testing CA and thus it can be concluded that FWA could be adopted as one of the new template algorithm for the training of ANNs.